🌌 AION-1: AstronomIcal Omnimodal Network

Polymathic's Large Omnimodal Model for Astronomy

🚀 Quick Start • 🎓 Tutorials • 🔬 Scientific Overview • 📦 Advanced Installation

---

🎯 Overview

AION-1 is a cutting-edge large omnimodal model specifically designed for astronomical surveys. It seamlessly integrates multiple data modalities, and enables simple adaptation to a wide range of astronomical tasks.

🚀 Quick Start

Assuming you have PyTorch installed, you can install AION trivially with:

pip install polymathic-aion

Then you can load the pretrained model and start analyzing astronomical data:

importtorchfromaionimportAIONfromaion.codecsimportCodecManagerfromaion.modalitiesimportLegacySurveyImage# Load model and codec managermodel=AION.from_pretrained('aion-base').to('cuda') # or 'aion-large', 'aion-xlarge'codec_manager=CodecManager(device='cuda')
# Prepare your astronomical data (example: Legacy Survey image)image=LegacySurveyImage(
flux=your_image_tensor, # Shape: [batch, 4, height, width] for g,r,i,z bandsbands=['DES-G', 'DES-R', 'DES-I', 'DES-Z']
)
# Encode data to tokenstokens=codec_manager.encode(image)
# Option 1: Extract embeddings for downstream tasksembeddings=model.encode(tokens, num_encoder_tokens=600)
# Option 2: Generate predictions (e.g., redshift)fromaion.modalitiesimportZpreds=model(
codec_manager.encode(image),
target_modality=Z,
)

🎓 Tutorials

Start with our interactive tutorial:

• Open in Google Colab - Learn AION basics interactively, no local setup required!

🔬 Scientific Overview

🧬 Architecture

AION-1 employs a two-stage, transformer-based design:

1. Modality-Specific Tokenizers transform raw inputs into discrete tokens
2. Unified Encoder–Decoder Transformer ingests all token streams via a multimodal masked modeling (4M) objective

---

🗂️ Supported Modalities

AION-1’s tokenizers cover 39 distinct data types, grouped by survey and data category

Category	Description	Token Name(s)
Imaging (2)	Legacy Survey, HSC Wide	tok_image_ls, tok_image_hsc
Catalog (1)	Legacy Survey catalog entries	catalog
Spectra (2)	SDSS, DESI	tok_spectrum_sdss, tok_spectrum_desi
Gaia (4)	BP/RP spectra, parallax, sky coords	tok_xp_bp, tok_xp_rp, tok_parallax, tok_ra, tok_dec
Gaia Photometry (3)	G/BP/RP flux	tok_flux_g_gaia, tok_flux_bp_gaia, tok_flux_rp_gaia
Legacy Survey (9)	g,r,i,z bands & WISE W1–W4 flux, E(B–V)	tok_flux_g,…,tok_flux_w4, tok_ebv
Legacy Shape (3)	Ellipticity components & effective radius	tok_shape_e1, tok_shape_e2, tok_shape_r
HSC Photometry (5)	g,r,i,z,y magnitudes	tok_mag_g,…,tok_mag_y
HSC Extinction (5)	g,r,i,z,y extinctions	tok_a_g,…,tok_a_y
HSC Shape (3)	Shape components 11,22,12	tok_shape11, tok_shape22, tok_shape12
Other (1)	Spectroscopic redshift	tok_z

---

📈 Model Variants

Variant	Encoder Blocks	Decoder Blocks	Model Dim	Heads	Total Params	Model
Base	12	12	768	12	300 M	aion-base
Large	24	24	1024	16	800 M	soon
XLarge	24	24	2048	32	3 B	soon

Pretraining – Global batch size: 8 192 – Steps: Base (1.5 days on 64 H100), Large (2.5 days on 100 H100), XLarge (3.5 days on 288 H100) – Optimizer: AdamW, peak LR 2 × 10⁻⁴, linear warmup + cosine decay

🔧 Data Preparation

AION uses a typed data system to understand the provenance of each astronomical observation. Each modality must be properly formatted:

Modality Types

fromaion.modalitiesimport (
LegacySurveyImage, HSCImage, # ImagesDESISpectrum, SDSSSpectrum, # SpectraLegacySurveyFluxG, HSCMagG, # PhotometryGaiaParallax, Z, # Scalars# ... and 30+ more modalities
)

Example: Preparing Legacy Survey Data

importtorchfromaion.modalitiesimportLegacySurveyImage, LegacySurveyFluxG# Format image data (shape: [batch, 4, height, width])image=LegacySurveyImage(
flux=torch.tensor(image_data, dtype=torch.float32),
bands=['DES-G', 'DES-R', 'DES-I', 'DES-Z']
)
# Format scalar photometryflux_g=LegacySurveyFluxG(value=torch.tensor([flux_values]))

Supported Data Formats

Survey	Modality	Required Format
Legacy Survey	Images	4-band (g,r,i,z), any resolution (auto-cropped to 96×96)
HSC	Images	5-band (g,r,i,z,y), any resolution
DESI/SDSS	Spectra	Flux, inverse variance, wavelength arrays
Gaia	BP/RP	Coefficient arrays (55 coefficients each)
All Surveys	Scalars	Single values or 1D tensors

---

💡 Example Use Cases

🔍 Similarity Search

Find galaxies similar to a query object across different modalities:

# Extract embeddings for similarity searchquery_embedding=model.encode(codec_manager.encode(query_image))
all_embeddings=model.encode(codec_manager.encode(*dataset_images))
# Find most similar objects using cosine similarityfromsklearn.metrics.pairwiseimportcosine_similaritysimilarity_scores=cosine_similarity(query_embedding, all_embeddings)
similar_objects=similarity_scores.argsort()[::-1][:10] # Top 10 similar

📊 Property Prediction

Build lightweight models on AION embeddings:

# Extract embeddings from multiple modalitiesembeddings=model.encode(codec_manager.encode(
image, spectrum, flux_g, flux_r, flux_i, flux_z
), num_encoder_tokens=900)
# Train simple regressor for stellar mass, redshift, etc.fromsklearn.neighborsimportKNeighborsRegressorregressor=KNeighborsRegressor(n_neighbors=5)
regressor.fit(embeddings.mean(axis=1), target_property)

🌌 Generative Modeling

Predict missing astronomical properties:

# Predict redshift from photometry + morphologypredictions=model(
codec_manager.encode(image, flux_g, flux_r, flux_i, flux_z),
target_mask={'tok_z': torch.zeros(batch_size, 1)},
num_encoder_tokens=600
)
redshift_probs=torch.softmax(predictions['tok_z'], dim=-1)

📦 Advanced Installation

AION offers flexible installation options to suit your environment and requirements.

To install AION with PyTorch included:

pip install polymathic-aion[torch]

For contributors and developers:

pip install polymathic-aion[torch,dev]

This includes testing frameworks, linting tools, and development dependencies.

For specific PyTorch versions (e.g., CUDA support):

# Install PyTorch with CUDA 12.4 support
pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https://download.pytorch.org/whl/cu124
# Then install AION
pip install polymathic-aion

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🌟 Acknowledgments

AION is developed by Polymathic AI, advancing the frontier of AI for scientific applications. We would like to acknowledge the support of the Simons Foundation and of Schmidt Sciences. This project was provided with computer and storage resources by GENCI at IDRIS thanks to the grant 2024-GC011015468 on the supercomputer Jean Zay’s H100 partition. Additionally, some of the computations in this work were run at facilities supported by the Scientific Computing Core at the Flatiron Institute, a division of the Simons Foundation.

📬 Contact

• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

_{Built with ❤️ for the astronomical community}